I am an AI Freeloader

The landscape just to start

I am reminded of the glut of steaming(+) services. Where cutting cable ending up seeming like a bad idea if you just wanted to watch the shows you loved. I have neither the time or money to waste on all of these.

…but I still want to use and learn all of them

So here’s what I do. I use all of their free tiers, evaluation models, loss leaders, etc. And while that sounds like a mess of context switching, and it can be, I have found it is entirely manageable with some careful planning.

A system for the chaos

I use two things to keep everythng in sync:

1. The Memory Bank prompt.
2. A script to sync all the environments.

The Memory Bank

I have a prompt that I use in every project. It is a markdown file that I keep in the project directory under prompts/instructions.md. It changes now and then, but unless I add something that turns out to be a real game changer I don’t ever really update projects using older version. I just start fresh with the new prompt in new projects. Here is my latest revision of this prompt: Memory Bank

The Sync Script

I use a script to sync all the environments. I am usually working in Node so it’s a .mjs script that I keep in the project directory under scripts/sync-ai.mjs. It could easily be ported to a shell script. Here is my latest revision of this script: Sync Script. All it is doing is sym linking the memory bank prompt into the various AI environments. Claude wants CLAUDE.MD, Cursor wants .cursor/instructions.md, etc.

Credit

I invented neither of these ideas. I don’t have a particular source to reference, i think the memory bank was a Cline thing at some point and the sync script is based on someone else’s idea I saw on Reddit. I just put them together in a way that works for me and have refined them for a while now.

The Workflow

1. Create a new project directory.
2. Copy the prompts/instructions.md and scripts/sync-ai.mjs files into the new project directory.
3. Run node scripts/sync-ai.mjs to sync all the environments.
4. Start using the whatever CLI or IDE you want to use.

➜  demo-ai-sync node scripts/sync-ai.mjs

=== Results ===

  ✅ GitHub Copilot: .github/copilot-instructions.md
  ✅ Windsurf: .windsurf/rules/instructions.md
  ✅ Anti Gravity: .agent/rules/instructions.md
  ✅ Claude Code: CLAUDE.MD
  ✅ Cursor: .cursor/rules/instructions.md
  ✅ OpenCode: AGENTS.MD

Synced to 6/6 platform(s)

➜  demo-ai-sync tree -L 3 -a
.
├── .agent
│   └── rules
│       └── instructions.md
├── .cursor
│   └── rules
│       └── instructions.md
├── .github
│   └── copilot-instructions.md
├── .windsurf
│   └── rules
│       └── instructions.md
├── AGENTS.MD
├── CLAUDE.MD
├── prompts
│   └── instructions.md
└── scripts
    └── sync-ai.mjs

10 directories, 8 files

The snags

Sometimes the AI’s need to be reminded of the instructions. I usually just start a new chat session with @instructions.md (or whatever the equivalent is for the tool). And after a completing a task if the AI has not update the memory bank I simply type “update memory bank” and it will update it. When I hit a limit the next AI will pick up where the previous one left off.

The cost

This is all free. I am not paying for any of these services. I am using the free tiers, evaluation models, loss leaders, etc. and the script + memory bank keeps them all on the same page.

The results

I am able to use all of these tools and get a good sense of what they can do. I am not locked into any one ecosystem. I can switch between them as needed. And I am not paying for any of them. It is a win win win.